Day 4

Detailed paper information

Back to list

Paper title Scaling preharvest phenotypic prediction of durum wheat grain quality and yield with UAV multispectral imaging for precision agriculture applications
  1. Shawn Carlisle Kefauver University of Barcelona, Faculty of Biology, Plant Physiology Section Speaker
  2. Joel Segarra University of Barcelona
  3. Thomas Vatter University of Barcelona, Faculty of Biology, Plant Physiology Section
  4. Ma. Luisa Buchaillot University of Barcelona
  5. Fatima Zahra Rezzouk University of Barcelona, Faculty of Biology, Plant Physiology Section
  6. Adrian Gracia-Romero University of Barcelona, Faculty of Biology, Plant Physiology Section
  7. Ruth Sanchez-Bragado University of Barcelona, Faculty of Biology, Plant Physiology Section
  8. María Teresa Nieto-Taladriz INIA-CSIC (Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria)
  9. Nieves Aparicio Gutiérrez Technological and Agrarian Institute of Castilla y León (ITACyL)
  10. Iker Aranjuelo Instituto de Agrobiotecnología (IdAB), CSIC- Gobierno de Navarra
  11. Ma. Dolores Serret Molins University of Barcelona, Faculty of Biology, Plant Physiology Section
  12. Jordi Bort Pie University of Barcelona, Faculty of Biology, Plant Physiology Section
  13. José Luis Araus Araus University of Barcelona, Faculty of Biology, Plant Physiology Section
Form of presentation Poster
  • C4. HAPs/UAVs
    • C4.01 Innovative UAV applications
Abstract text In addition to crop productivity, food quality traits are of high importance for farmers and a major factor affecting end-use product quality and human health. Food quality has been specifically identified among the United Nations Sustainable Development Goals (SDGs) as a key component of Goal 2 Zero Hunger, to end hunger in part through improved nutrition. Durum wheat is one of the most important cereal grains grown in the Mediterranean basin where the strong influence of climatic change complicates agricultural management and efforts to develop environmentally adapted varieties with higher yields and also improved quality traits. Protein content is among the most important wheat quality features, nonetheless in the last decades a reduction in durum wheat protein content has been observed associated with the spread of high yielding varieties. Therefore, it is central to develop efficient quality-related phenotyping and monitoring tools. Predicting not only yield but also important quality traits like protein content, vitreousness, and test weight in the field before harvest is of high value for breeders aiming to optimize crop resource allocation and develop more resilient crops. Moreover, the relation between grain protein and nitrogen fertilization plays a central role in the sustainability of agriculture management, again connecting these efforts to the SDG 2.
In this study, we take a two-pronged approach towards improving both yield quantity and grain quality estimations of durum wheat across Spain. With this aim in mind, we brought together the confluence of crop phenotyping and precision agriculture through incorporating genetic, environmental and crop management factors (GxExM) at multiple scales using different remote sensing approaches. Aiming to develop efficient phenotyping tools using remote sensing instruments and to improve field-level management for more efficient and sustainable monitoring of grain nitrogen status, the research presented here focuses on the efficacy of multispectral and high resolution visible red-green-blue (RGB) imaging sensors at different scales of observation and crop phenological stages (anthesis to grain filling).
Linear models were calculated using vegetation indices at each sensing level, sensor type and phenological stage for intercomparisons of sensor type and scale. Then, we used machine learning (ML) models to predict grain yield and important quality traits in crop phenotyping microplots using 11-band multispectral UAV image data. Combining the 11 multispectral bands (450 ± 40, 550 ± 10, 570 ± 10, 670 ± 10, 700 ± 10, 720 ± 10, 780 ± 10, 840 ± 10, 860 ± 10, 900 ± 20, 950 ± 40 nm) for 34 cultivars and 16 environments supported the development of robust ML models with good prediction capability for both yield and quality traits. Applying the trained models to test sets explained a considerable degree of phenotypic variance at good accuracy with R2 values of 0.84, 0.69, 0.64, and 0.61 and normalized root mean squared errors of 0.17, 0.07, 0.14, and 0.03 for grain yield, protein content, vitreousness, and test weight, respectively.
Following these findings, we modified our UAV multispectral sensor to match Sentinel-2 visible and near-infrared spectral data bands in order to better explore the upscaling capacities of the grain yield and protein linear models. Specifically, models built at anthesis with UAV multispectral red-edge band data performed best at grain nitrogen content estimation (R2=0.42, RMSE=0.18%), which can be linked to grain protein content. We also demonstrated the possibility to apply the UAV-derived phenotyping models to satellite data and predict grain nitrogen content for actual wheat fields (R2=0.40, RMSE=0.29%). Results of this study show that using ML models of multispectral UAV can be a powerful approach to efficiently predict important quality traits and yield preharvest at the micro-plot level in phenotyping trials. Furthermore, we demonstrate that phenotyping microplot-based grain quality and grain yield prediction models are amenable to Sentinel-2 satellite precision agriculture applications at larger scales, representing an effective synergy based on the inherent scalability of remote sensing for assessing plant physiological primary and secondary traits.